From ~24 months old, typically developing (TD) children often generalise names for solid objects to novel examples that are the same shape. This ‘shape bias’ can facilitate word learning by providing an attentional short-cut, allowing children to accurately generalise novel nouns. Autistic children often experience delays in language acquisition, and difficulty exploiting the shape bias may be a contributing factor. However, extant research findings are highly inconsistent, with autistic children exhibiting a shape bias in certain tasks but not others. Thus, the current research investigated whether variability in autistic children’s shape bias can be explained by methodological differences.

Autistic children (aged 4 to 9 years) and TD children matched on receptive vocabulary (aged 30 months to 4 years) participated in five experiments measuring shape bias in both ‘forced-choice’ and ‘yes or no’ variants of a novel noun generalisation task (NNG). Each task included an ‘online’ condition, where children were asked to generalise a label from a visible standard item to a target, and an ‘offline’ condition where the standard was absent at test, and generalisation had to be completed from memory.

Experiment 1 investigated whether the visibility of the standard affected autistic children’s shape bias in a forced-choice task with high contrast stimuli. We found that both autistic and TD groups generalised by shape regardless of standard visibility. Experiment 2 investigated whether shape was still preferred in a yes/no task, where children had the freedom to include any of the stimuli in the category rather than just the best example. In this task, autistic children were more likely to accept the differently shaped distractors than the TD group. Experiments 3 and 4 used the same tasks with low contrast stimuli to investigate whether the requirement to remember the standard had an impact when object shapes were more similar. Both groups again exhibited a shape bias in the forced-choice task, however in the yes/no version only the TD group generalised by shape. Finally, Experiment 5 investigated whether an attentional preference for small details could account for shape bias differences in autism in a yes/no task. We found that both autistic and TD children generalised based on global shape rather than a salient local feature.

Overall, our results suggest that methodological variations can explain discrepancies in previous findings regarding shape bias in autism. Autistic children exhibited a strong shape bias in forced-choice tasks, whereas the bias appeared reduced or absent in yes/no tasks requiring children to categorise items individually. This suggests that differences may lie in autistic children’s use of the shape bias as a tool for category exclusion decisions, rather than inclusion decisions, raising questions about the role of the shape bias in word and category learning for all children. There may be multiple routes through which attention to shape can contribute to learning and, by identifying which routes are most accessible for autistic children, we can inform teaching methods that work in harmony with their strengths.